用Predict.py进行推理时，无法使用GPU进行推理，一直默认使用CPU，请问如何解决？

[Qiuyanghang]

问题确认 Search before asking

• 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

刚接触paddleseg，配置好环境后想直接测试提供的预训练模型效果如何，但其始终用CPU进行推理，不清楚是什么原因，请问如何调用GPU推理。
尝试了以下操作都不管：1.用命令行参数设置GPU设备 2.pycharm终端运行前用 set CUDA_VISIBLE_DEVICES=0 设置GPU
python版本：3.9 paddlepaddle-gpu版本： 2.6.1.post116 paddleseg版本：2.8.0 cuda版本：11.6
window系统，一张3060显卡。
预测脚本运行时的信息：
------------Environment Information-------------
platform: Windows-10-10.0.19045-SP0
Python: 3.9.19 (main, Mar 21 2024, 17:21:27) [MSC v.1916 64 bit (AMD64)]
Paddle compiled with cuda: True
NVCC: Build cuda_11.6.r11.6/compiler.30794723_0
cudnn: 8.6
GPUs used: 1
CUDA_VISIBLE_DEVICES: None
GPU: ['GPU 0: NVIDIA GeForce']
PaddleSeg: 2.8.0
PaddlePaddle: 2.6.1
OpenCV: 4.5.5

2024-05-29 19:05:57 [INFO]
---------------Config Information---------------
batch_size: 4
iters: 160000
train_dataset:
dataset_root: data/cityscapes
mode: train
transforms:

• max_scale_factor: 2.0
  min_scale_factor: 0.5
  scale_step_size: 0.25
  type: ResizeStepScaling
• crop_size:
  • 1024
  • 512
    type: RandomPaddingCrop
• type: RandomHorizontalFlip
• brightness_range: 0.5
  contrast_range: 0.5
  saturation_range: 0.5
  type: RandomDistort
• type: Normalize
  type: Cityscapes
  val_dataset:
  dataset_root: data/cityscapes
  mode: val
  transforms:
• type: Normalize
  type: Cityscapes
  optimizer:
  momentum: 0.9
  type: SGD
  weight_decay: 0.0005
  lr_scheduler:
  end_lr: 0
  learning_rate: 0.005
  power: 0.9
  type: PolynomialDecay
  warmup_iters: 1000
  warmup_start_lr: 1.0e-05
  loss:
  coef:
• 1
• 1
• 1
  types:
• min_kept: 130000
  type: OhemCrossEntropyLoss
• min_kept: 130000
  type: OhemCrossEntropyLoss
• min_kept: 130000
  type: OhemCrossEntropyLoss
  model:
  backbone:
  pretrained: https://bj.bcebos.com/paddleseg/dygraph/PP_STDCNet2.tar.gz
  type: STDC2
  type: PPLiteSeg
  test_config:
  aug_eval: true
  scales: 1.0

---

2024-05-29 19:05:57 [INFO] Set device: cpu
2024-05-29 19:05:57 [WARNING] Add the num_classes in train_dataset class to model config. We suggest you manually set num_classes in model config.
2024-05-29 19:05:57 [WARNING] Add the in_channels in train_dataset class to model config. We suggest you manually set in_channels in model config.
2024-05-29 19:05:57 [INFO] Use the following config to build model
model:
backbone:
in_channels: 3
pretrained: https://bj.bcebos.com/paddleseg/dygraph/PP_STDCNet2.tar.gz
type: STDC2
num_classes: 19
type: PPLiteSeg
2024-05-29 19:05:58 [INFO] Loading pretrained model from https://bj.bcebos.com/paddleseg/dygraph/PP_STDCNet2.tar.gz
2024-05-29 19:05:58 [INFO] There are 265/265 variables loaded into STDCNet.
2024-05-29 19:05:58 [INFO] The number of images: 33
2024-05-29 19:05:58 [INFO] Loading pretrained model from my_model/model_pp.pdparams
2024-05-29 19:05:59 [WARNING] ppseg_head.arm_list.0._scale is not in pretrained model
2024-05-29 19:05:59 [WARNING] ppseg_head.arm_list.1._scale is not in pretrained model
2024-05-29 19:05:59 [WARNING] ppseg_head.arm_list.2._scale is not in pretrained model
2024-05-29 19:05:59 [INFO] There are 367/370 variables loaded into PPLiteSeg.
2024-05-29 19:05:59 [INFO] Start to predict...
7/33 [=====>........................] - ETA: 58s Traceback (most recent call last):